Hockey is a very complex sport with many different, unique facets to the game. One can imagine that in the NHL, where all 30 teams are constantly battling for the chance to win the Stanley Cup, each club is looking for any sort of advantage it can gain, with no area of the game being left untouched. And while the majority of the game is played at even-strength, a significant portion is still played on the power play, where a team is given the advantage of having one more player on the ice than the opposing team after a penalty. Each team has its own tactics and strategies for their power plays but there is at least one crucial decision every team must make when constructing a power play game plan. A team has the option of sending out a unit compromised of three forwards and two defenseman or four forwards and only one defenseman, with a seemingly slight preference towards the former. But which is better, using three forwards or four on a power play?
While not a lot of formal statistical research has been conducted on specific aspects of the power play and penalty kill in the NHL, there is some precedent in analyzing the effect of the number of forwards deployed with the man advantage. Matt Cane has done the most research, digging into the differences between three forward power play units and four forward power play units over at Hockey-Graphs.com. In his research, he hypothesized that the benefits of utilizing an extra forward on the power play would make itself known in two specific aspects: there would most likely be an increase in shot generation (Corsi For) and an increase in shooting percentage. Gathering data from the past six years, Cane ended up seeing exactly that. Compared to three forward power play units, there was an uptick in both shot generation and shot conversion by the four forward units. Cane states that “the advantage to the 4-forward approach here is pretty clear – the increase in Corsi For works out to more than 10 extra shot attempts per 60 minutes of powerplay time, or about an extra shot attempt every 3 powerplays, and on top of that each shot on goal goes in at a significantly higher rate” (Cane para 6).
Cane would also go on to look at each unit’s goal differential and found similar results, with the four forward units having a goal differential of 6.1 per 60 minutes while the three forward units only had a goal differential of 4.9. Overall, his research suggests that a four forward power play unit is more effective than one that only deploys three forwards. This would align with a person’s initial thoughts on the subject, as forwards are inherently better offensively and more skilled at scoring goals than defenseman, so having more forwards on the ice should most likely lead to more offense.
There will be five statistical tests conducted so there will be five null hypotheses and five alternative hypotheses, one for each test. Each test has an alpha value equal to 0.05, where ⍺ is the significance level. For the only two prob z-test, the null hypothesis, or Ho, is that the true proportion of successful zone entries by four forward units is equal to the true proportion of successful zone entries by three forward units. The alternative hypothesis, or Ha, is that the true proportion of successful zone entries by four forward units is greater than the true proportion of successful zone entries by three forward units.
The remaining four statistical tests will all be two sample t-tests, measuring the difference between a four forward unit and three forward unit’s mean goals, shots, shot attempts, or scoring chances. All of the null hypotheses will be that the true mean number of goals, shots, shot attempts, or scoring chances of four forward power play units is equal to the true mean number of goals, shots, shot attempts, or scoring chances of three forward power play units. The corresponding alternative hypotheses for these tests will be that the four forward unit’s true mean goals, shots, shot attempts, or scoring chances is greater than the three forward unit’s.
To try and determine which power play set up is better, I collected my own data from NHL games to see how each unit performed in real matches. During this year’s NHL playoffs, I individually tracked games from the postseason and analyzed how the offensive team’s power play performed, in addition to how many forwards they deployed during the attack. Each individual entry in the data was logged as each attempt to enter in the offensive zone, not just as the power play as a whole.
In total, there were 957 individual entries. 520 entries were attempted with four forwards while 437 entries were attempted using three forwards. All entries originated in games from the Stanley Cup Playoffs so there would not be a significant difference in talent between two teams in an individual game. In addition, every team in the postseason was tracked at least once.
Before I began the data collection, I first had to define what makes a power play effective and what exactly I would be tracking. Then once I collected enough data, I could analyze it and see which power play unit performed better. Obviously the chief objective of a power play is to score, but the problem with analyzing only goals is that there is a relatively small number of them and they do not predict future goals relatively well. So in addition to goals, to define which power play is more effective I looked at a number of other factors that have been proven to be very important in scoring goals in the NHL. These factors (such as shots, shot attempts, and scoring chances) have all been shown to correlate well with goals and are relatively very repeatable measures. From each possession in the offensive zone, I tracked how many shots, shot attempts, and scoring chances the attacking team generated. Even though scoring chances are not an official statistic tracked by the National Hockey League, I used the conventional method of categorizing them. Any shot attempt that was taken inside the “home-plate” area in front of the net was marked down as a scoring chance, as it is usually recorded as.
Even though there were around 1,000 entries in total tracked, sample size could be a bit of a concern, especially considering the data only comes from one postseason. Collecting data over multiple seasons would be idea, but it was not feasible given the time and resource limits. Each entry was manually tracked and very labor intensive and around 1,000 entries should still provide a representative sample.
However, the integrity of the data should not be called into question as it was checked against the official National Hockey League Play By Play data. While human error is always a possibility, all entries should be accurate as to what corresponded during the games.
In my tests for goals, shots, shot attempts, scoring chances, and success percentage, I found p-values of .43, .079, .099, .067, and .091 respectively. At an alpha level of .05, one fails to reject the null hypotheses in all the tests, displaying no evidence to support the alternative hypotheses. Thus, it can be assumed that the true mean goals, shots, shot attempts, and scoring chances of four forward units are equal to the true mean goals, shots, shot attempts, and scoring chances of three forward units and the true proportion of successful entries of four forward units is equal to the true proportion of successful entries by three forward units. Since the sample should be representative of the population, these findings are generalizable to the entire NHL.
The study was designed to determine if a four forward power play unit was a better option than a three forward power play unit in the NHL by investigating whether or not the four forward unit significantly outperformed the three forward unit at a .05 alpha value, at which one rejects the null hypothesis. While at the .05 alpha value one fails to reject the null hypothesis, at the .1 alpha value one can reject the null hypotheses that the true mean number of shots, scoring chances, or shot attempts by a four forward unit are equal to the true mean number of shots, scoring chances, or shot attempts by a three forward unit. By rejecting these null hypotheses, there is evidence to support the claim that the true mean number of shots, scoring chances, or shot attempts by a four forward unit are greater than the true mean number of shots, scoring chances, or shot attempts by a three forward unit. From this, one can assume that there are slight benefits to a four forward unit compared to a three forward unit. Collecting more data and gathering a larger sample size could help determine if these findings are representative of the true effectiveness of both types of power play units.
*In the next few days, I plan on writing another piece about my findings regarding four power play units and three power play units, one that is less math-inclined and more focused on the results*